Construction and validation of a prediction model for 90-day readmission risk in patients with chronic heart failure

TL;DR

This study creates a model to predict which patients with chronic heart failure are at high risk of being readmitted within 90 days, using factors like blood test results and glucose levels.

Contribution

The study introduces a novel prediction model for 90-day readmission risk in CHF patients using LASSO regression and clinical biomarkers.

Findings

A predictive model with AUC of 0.746 in training and 0.705 in validation cohorts was developed for 90-day CHF readmission.

Five biomarkers (cTn, FBG, serum sodium, eGFR, NEU) were identified as key predictors of readmission risk.

The model showed good discriminative power and clinical validity for guiding personalized management.

Abstract

Chronic heart failure (CHF) is associated with high morbidity and mortality rates, which is not curable currently, resulting in an increasing risk of readmission and imposing a considerable burden on healthcare systems. Predictive modeling is a critical tool for guiding the clinical management of CHF. 90-day is a crucial time point for readmission risk assessment in patients with CHF. However, there is a lack of risk factor exploration, as well as predictive modeling for 90-day readmission risk in these patients. The aim of this study is to identify prognostic risk biomarkers and develop a novel prediction model for 90-day readmission for patients with CHF.

542 CHF patients hospitalized at the Department of Cardiology, the Fourth Affiliated Hospital of Zhejiang University were randomly split into training (N = 380) and validation (N = 162) cohort at a 7:3 ratio. Demographic, comorbidities, laboratory tests, and echocardiography results were analyzed through Least Absolute Shrinkage and Selection Operator (LASSO) regression to select predictive variables. Furthermore, receiver operating characteristic (ROC) curve, the area under the curve (AUC), decision curve analysis (DCA), and calibration curves were used to access the discriminative power, clinical validities, and calibration of the model.

Of the included 542 patients, the readmission rates were 18.7% and 19.1% in 90-day follow-up in the training and validation cohort respectively. Five variables, including cardiac troponin (cTn), fasting blood glucose (FBG), serum sodium, estimated glomerular filtration rate (eGFR), neutrophil (NEU) showed the strongest correlation with 90-day readmission according to LASSO regression. These selected variables were then combined into a novel prediction model, with an AUC of 0.746 [95% (confidence interval) CI: 0.685–0.808] in the training cohort and 0.705 (95% CI: 0.605–0.804) in the validation cohort.

Our findings suggest that a predictive model incorporating the variables of cTn, FBG, serum sodium, eGFR and NEU demonstrating a good predictive ability for 90-day readmission risk in patients with CHF, which can aid clinicians in clinical decisions and personalized management.

Figures